From: Matthias Braun Date: Tue, 20 Mar 2007 12:58:30 +0000 (+0000) Subject: improve some bitfiddle functions, add ceil_po2 and floor_po2 X-Git-Url: http://nsz.repo.hu/git/?a=commitdiff_plain;h=f3190ef63f207a867df82a22b7e72b5ec197bc1b;p=libfirm improve some bitfiddle functions, add ceil_po2 and floor_po2 [r8720] --- diff --git a/ir/adt/bitfiddle.h b/ir/adt/bitfiddle.h index e01eea3ac..624a2d677 100644 --- a/ir/adt/bitfiddle.h +++ b/ir/adt/bitfiddle.h @@ -1,20 +1,20 @@ /** - * @file bitfiddle.h - * @date 28.9.2004 - * @brief Functions from hackers delight. - * - * Attention! These functions silently assume, that an int is 32 bit wide. - * $Id$ + * @file + * @date 28.9.2004 + * @brief Functions from hackers delight. + * @author Sebastian Hack, Matthias Braun + * @version $Id$ */ - -#ifndef __FIRM_HACKDEL_H -#define __FIRM_HACKDEL_H +#ifndef _FIRM_BITFIDDLE_H_ +#define _FIRM_BITFIDDLE_H_ #include +#include "util.h" -#include "firm_config.h" - +/* some functions here assume ints are 32 bit wide */ #define HACKDEL_WORDSIZE 32 +COMPILETIME_ASSERT(sizeof(unsigned) == 4, unsignedsize) +COMPILETIME_ASSERT(UINT_MAX == 4294967295U, uintmax) /** * Add saturated. @@ -24,7 +24,8 @@ * * @note See hacker's delight, page 27. */ -static INLINE int add_saturated(int x, int y) +static inline __attribute__((const)) +int add_saturated(int x, int y) { int sum = x + y; /* @@ -37,10 +38,12 @@ static INLINE int add_saturated(int x, int y) /* The infinity to use. - Make a mask of the sign bit of x and y (they are the same if an overflow occurred). - INT_MIN == ~INT_MAX, so if the sign was negative, INT_MAX becomes INT_MIN. + Make a mask of the sign bit of x and y (they are the same if an + overflow occurred). + INT_MIN == ~INT_MAX, so if the sign was negative, INT_MAX becomes + INT_MIN. */ - int inf = (x >> (sizeof(x) * 8 - 1)) ^ INT_MAX; + int inf = (x >> (sizeof(x) * 8 - 1)) ^ INT_MAX; return overflow < 0 ? inf : sum; } @@ -50,13 +53,14 @@ static INLINE int add_saturated(int x, int y) * @param x A 32-bit word. * @return The number of bits set in x. */ -static INLINE unsigned popcnt(unsigned x) { - x = x - ((x >> 1) & 0x55555555); - x = (x & 0x33333333) + ((x >> 2) & 0x33333333); - x = (x + (x >> 4)) & 0x0f0f0f0f; - x = x + (x >> 8); - x = x + (x >> 16); - return x & 0x3f; +static inline __attribute__((const)) +unsigned popcnt(unsigned x) { + x -= ((x >> 1) & 0x55555555); + x = (x & 0x33333333) + ((x >> 2) & 0x33333333); + x = (x + (x >> 4)) & 0x0f0f0f0f; + x += x >> 8; + x += x >> 16; + return x & 0x3f; } /** @@ -64,13 +68,25 @@ static INLINE unsigned popcnt(unsigned x) { * @param x The word. * @return The number of leading (from the most significant bit) zeros. */ -static INLINE unsigned nlz(unsigned x) { - x |= x >> 1; - x |= x >> 2; - x |= x >> 4; - x |= x >> 8; - x |= x >> 16; - return popcnt(~x); +static inline __attribute__((const)) +unsigned nlz(unsigned x) { +#ifdef USE_X86_ASSEMBLY + unsigned res; + if(x == 0) + return 32; + + __asm__("bsrl %1,%0" + : "=r" (res) + : "r" (x)); + return 31 - res; +#else + x |= x >> 1; + x |= x >> 2; + x |= x >> 4; + x |= x >> 8; + x |= x >> 16; + return popcnt(~x); +#endif } /** @@ -78,7 +94,21 @@ static INLINE unsigned nlz(unsigned x) { * @param x The word. * @return The number of trailing zeros. */ -#define ntz(x) (HACKDEL_WORDSIZE - nlz(~(x) & ((x) - 1))) +static inline __attribute__((const)) +unsigned ntz(unsigned x) { +#ifdef USE_X86_ASSEMBLY + unsigned res; + if(x == 0) + return 32; + + __asm__("bsfl %1,%0" + : "=r" (res) + : "r" (x)); + return res; +#else + return HACKDEL_WORDSIZE - nlz(~x & (x - 1)); +#endif +} /** * Compute the greatest power of 2 smaller or equal to a value. @@ -104,6 +134,61 @@ static INLINE unsigned nlz(unsigned x) { */ #define round_up2(x,pot) (((x) + ((pot) - 1)) & (~((pot) - 1))) +/** + * Returns the biggest power of 2 that is equal or smaller than @p x + * (see hackers delight power-of-2 boundaries, page 48) + */ +static inline __attribute__((const)) +unsigned floor_po2(unsigned x) +{ +#ifdef USE_X86_ASSEMBLY // in this case nlz is fast + if(x == 0) + return 0; + // note that x != 0 here, so nlz(x) < 32! + return 0x80000000U >> nlz(x); +#else + x |= x >> 1; + x |= x >> 2; + x |= x >> 4; + x |= x >> 8; + x |= x >> 16; + return x - (x >> 1); +#endif +} +/** + * Returns the smallest power of 2 that is equal or greater than x + * @remark x has to be <= 0x8000000 of course + * @note see hackers delight power-of-2 boundaries, page 48 + */ +static inline __attribute__((const)) +unsigned ceil_po2(unsigned x) +{ + if(x == 0) + return 0; + assert(x < (1U << 31)); + +#ifdef USE_X86_ASSEMBLY // in this case nlz is fast + // note that x != 0 here! + return 0x80000000U >> (nlz(x-1) - 1); +#else + x = x - 1; + x |= x >> 1; + x |= x >> 2; + x |= x >> 4; + x |= x >> 8; + x |= x >> 16; + return x + 1; +#endif +} + +/** + * Tests whether @p x is a power of 2 + */ +static inline __attribute__((const)) +int is_po2(unsigned x) +{ + return (x & (x-1)) == 0; +} #endif